Chris Blattman

Statistics bleg

Imagine you’re a microeconomist moderately skilled in survey and panel data analysis, but wholly ignorant of time series. You have vague recollections of learning what a VAR was five years ago. (You might also be charming and handsome, but that could be self-delusion speaking.)

One day you wake up and think you want to estimate impulse response functions using cross-country time series data. What do you read that (1) gives you the basic technique, or (2) illustrates an applied example clearly?

Assume moderate ignorance.

In case it helps, it so happens that you want to look at the political response (mainly binary events) to exogeneous price shocks. You have both monthly and annual data, painstakingly collected in the eighth basement of Harvard and Berkeley libraries as a (admittedly pale) grad student, before you discovered the delights of field work.

Hypothetically-speaking, of course.

11 Responses

  1. Not sure that this is pertinent any more, but I would suggest taking a look at David Hendry’s “Dynamic Econometrics”, which is a great text insofar as he really helps illustrate the fundamental assumptions about the data generating process that give rise to different models.

    As far as implementing standard time series models, Helmut Lutkepohl’s book(s) are great references. They make it comparatively straightforward to code up your own program in something like Matlab.

    Anyhow, best of luck. Sounds fun.

  2. Many things can go wrong when setting up a VAR, and impulse response are fairly sensitive to identification and restrictions. Setting up a error correction model is far from trivial even if some packages like Eviews or RATS/CATS can give you that feeling. My two cents: work with a time series guy. Isn’t division of labour and specialization one of the few things we hold dear?

  3. Kennedy’s Guide to Econometrics is the first place I look at for econometrics.

    @Fernando, Suzanna de Boef and Luke Keele showed in an article (“Taking Time Seriously”, AJPS 2008) that EC models can be used with stationary data. they are not limited to cointegrated time series.

    1. That is correct – with stationary data the ECM is a restricted ADL model.

      However, typically these models are used when data is not stationary, for that is when they have the greatest leverage. (Not sure what happens to the super consistency property of ECM in the context of stationary data)

      In cases of non-stationarity, testing for co-integration is necessary but, in my experience, not always done.

  4. Stock and Watson’s Introduction to Economerics is the place to begin – first edition is better than the second edition.

  5. I first learned about VARs from reading the RATS User’s Guide, and then I used Enders’ book and Hamilton’s book as a reference as needed. But I felt the RATS User’s Guide had a very accessible introduction (see chapter 10 here)

  6. This paper by Love and Zecchino has something (but not much). At some point I had seen a pvar command for STATA written by the authors, but I don’t think it is available on line. Christian Broda has a nice application in “Terms of Trade and Exchange Rate Regimes in Developing Countries.” Journal of International Economics Vol. 63, pp. 31-58, May 2004.
    However, I’ve never seen a paper with a nice explanation of panel var, so if you find out let us know.

  7. Chris,
    Not that I have a lot of knowledge about econometrics but I do have a very good foundation so I will strongly recommend that you read:
    1. Walter Enders “Applied Econometric Time Series”. This will set you up for some basic knowledge on time series matters. It also has some very good examples that can be pretty useful mathematically and … well for life somehow.
    2. I don’t know how good this may sound but we used to practice in Eviews. Now this program is obviously not very popular among researchers but I have to say that, for somebody that is starting or refreshing his knowledge in time series, it may be a pretty friendly tool for practicing. Also, its help guide includes two things: instructions of how to apply the commands and use this program for specific estimations AND some econometric sense and theory of what you may be doing.
    3. If you feel this book and Eviews are way too basic, I would strongly recommend the James D. Hamilton “Time Series Analysis”. In an personal opinion, it is too “dark” for my taste but, I’ve found friends (who are much more advaced than me in this “econometric world”) who really will describe it as their basic textbook when it comes to time series analysis.

    Hope that helps!

  8. 1. Visit Prof. Neal Beck’s website and read his co-authored papers in time series, binary time series cross section, etc.

    2. Read Walter Enders “Applied Econometric Time Series” – Although deals mainly with univariate time series, is good to understand some intricacies of stationarity, seasonality, etc.

    3. More references for short and long panels in Greene’s and Baltagi’s latest textbooks, both of which touch on the subject.

    4. Finally, a warning. Just because we can all write an error correction model, that does not make the process that generated the data error correction. For a process to be EC you obviously need to estimate an EC model, but also ensure the series (1) are integrated of the same order and, if so, (2) co-integrated (e.g. simply put, the residuals of the long term relation are stationary). Unfortunately to many papers ignore 1 and 2.

  9. Should you receive a response by email, please share it on your blog. I have this same question about time series. Thanks

Comments are closed.

Why We Fight - Book Cover
Subscribe to Blog